Академический Документы
Профессиональный Документы
Культура Документы
Lecture I, UL 2007
Finbarr Holland,
Department of Mathematics,
University College Cork,
f.holland@ucc.ie;
July 12, 2007
Three Problems
a1
a2
a3
=
= .
b1
b2
b3
1
b1
1
c1
1
1
a2
1
b2
1
c2
1
1
a3
1
b3
1
c3
1
1
a1 +a2 +a3
a1
a2
a3
=
= .
b1
b2
b3
A basic inequality
b
2 ab ta + , t > 0,
t
1
1
b1 +b2 +b3
1
c1 +c2 +c3
r
t=
b
.
a
In other words,
b
2 ab = min{ta + : t > 0} = min{sa + bt : s, t > 0, st = 1}.
t
Proof.
b
at2 2t ab + b
(t a b)2
ta + 2 ab =
=
.
t
t
t
We deduce some consequences.
Theorem 2. If a, b, c, d > 0. Then
p
ab + cd (a + c)(b + d),
with equality iff
a
b
= .
c
d
b
d
(b + d)
2[ ab + cd] at + + ct + = (a + c)t +
,
t
t
t
with equality iff
r
r
r
b
d
b+d
t=
=
=
.
a
c
a+c
Hence
p
(b + d)
2[ ab + cd] min{(a + c)t +
: t > 0} = 2 (a + c)(b + d).
t
We can build on this in two ways.
Theorem 3. Suppose a, b, c, d, e, f, g, h > 0. Then
p
p
4
abcd + 4 ef gh 4 (a + e)(b + f )(c + g)(d + h),
with equality iff
a
b
c
d
= = = .
e
f
g
h
4
abcd + 4 ef gh =
xy + uv
p
(x + u)(y + v)
q
p
p
( ab + ef )( cd + gh)
=
qp
p
(a + e)(b + f ) c + g)(d + h)
p
= 4 (a + e)(b + f )(c + g)(d + h).
We can now deduce the following statement.
2
3
abc + 3 ef g 3 (a + e)(b + f )(c + g),
with equality iff
Proof. Let d =
whence
a
b
c
= = .
e
f
g
abc, h =
4
d+h =
abcd + 4 ef gh
p
4 (a + e)(b + f )(c + g)(d + h),
abc +
p
p
3
ef g = d + h 3 (a + e)(b + f )(c + g).
n
a1 a2 an + n b1 b2 bn n (a1 + b1 )(a2 + b2 ) (an + bn ),
with equality iff
a1
a2
an
=
= = .
b1
b2
bn
Theorem 2 is a special case of a disguised version of one form of perhaps the most
useful inequality in all of Mathematics, namely the Cauchy-Schwartz inequality.
Theorem 4 (Cauchy-Schwartz). If xi , yi > 0, i = 1, 2, . . . , n, then
v
u n
n
n
X
X
uX
t
xi yi (
xi )(
yi ),
i=1
i=1
i=1
x2
xn
x1
=
= =
.
y1
y2
yn
Proof. Invoke Theorem 2 and use induction. We can phrase this differently.
Corollary 2. If ai , i = 1, 2, . . . , n are real numbers, then
v
u n
n
n
X
X
uX
2
t
x2i = 1}.
ai xi | :
ai = max{|
i=1
i=1
i=1
Proof. For, if
Pn
i=1
x2i = 1, then
|
n
X
ai xi |
i=1
n
X
i=1
n
X
|ai | |xi |
p
(ai )2 (xi )2
i=1
v
u n
n
X
uX
t(
a2i )(
x2i )
i=1
v
u n
uX
= t
a2i ,
i=1
i=1
a2i
, i = 1, 2, . . . , n.
(a + b)2 + (c + d)2 a2 + c2 + b2 + d2 .
Proof. Suppose x2 + y 2 = 1. Then
a2 + c2 +
b2 + d2 .
It follows that
p
i=1
Proof. Suppose
i=1
i=1
Pn
x2i = 1. Then
v
v
u n
u n
n
n
X
X
uX
uX
2
t
ai + t
b2i |
a i xi | + |
b i xi |
i=1
i=1
i=1
i=1
(ai + bi )xi |.
i=1
i=1
n
X
Hence
v
v
v
u n
u n
u n
n
n
X
X
X
X
u
u
uX
2
2
t (ai + bi )2 = max{|
t
(ai + bi )xi | :
xi = 1}
ai + t
b2i .
i=1
i=1
i=1
i=1
i=1
Discussion of Theorem 2
g2 (x) = x1 x2 .
Then, if y = (y1 , y2 ),
g2 (x) + g2 (y) g2 (x + y).
Thus g2 is super-additive on R2 .
Notice, too, that, for every scalar > 0, x = (x1 , x2 ) and
g2 (x) = g2 (x), x R2 ,
i.e., g2 is homogeneous of degree 1. Hence, g2 is concave on R2 . Analytically, this
means that x, y R2 , and 0 1, then
g2 (x) + (1 )g2 (y) g2 (x + (1 )y).
To get an idea of what it means, geometrically, try and picture the surface in 3-space
whose equation is
z = xy, x, y 0.
Exercise 4. Apropos Problem 2, draw a similar conclusion about the function
g3 (x) = 3 x1 x2 x3 , x = (x1 , x2 , x3 ) R3 , xi 0, i = 1, 2, 3.
In fact, if x = (x1 , x2 , . . . , xn ) is a vector in n-space and its coordinates are positive,
and
gn (x) = n x1 x2 xn ,
then gn is super-additive and is homogeneous of degree 1 on
{x Rn : xi 0, i = 1, 2, . . . , n}.
Hence, gn is concave on this region.
5
Discussion of Theorem 5
Its familiar that a2 + b2 is the euclidean distance from the origin to the point
(a, b) in R2 . Thus Theorem 5 is simply an analytical statement equivalent to the
triangle inequality! To put this another way, let
q
`2 (x) = x21 + x22 , x = (x1 , x2 ) R2 ,
then
`2 (x + y) `2 (x) + `2 (y), x, y R2 .
In other words, `2 is sub-additive. But its also homogeneous of degree one. Hence,
it is convex on R2 , i.e., x, y R2 , and 0 1, then
`2 (x + (1 )y) `2 (x) + (1 )`2 (y).
To get an idea of what this means, geometrically, try and picture the region of points
(x, y, z) in 3-space constrained by the conditions
p
0 z x2 + y 2 , x, y R.
[Hint: at your next meal examine the sugar bowl, say.]
A solution of Problem 1 follows once we can show that
q
`3 (x) = x21 + x22 + x23 , x = (x1 , x2 , .x3 ) R3 ,
has the same property.
Exercise 5. Establish that `3 is sub-additive and convex on R3 , and solve Problem
1.
Solution of Problem 3
By now it will have occurred to you that what is wanted to solve Problem 3 is an
analogue of Corollary 1 for the harmonic mean of three positive numbers. (This has
properties similar to the geometric mean, but doesnt get as much attention, and
doesnt feature in too many IMO problems.) Namely, we want to show that the
function h3 defined by
h3 (x) =
3
1
x1
1
x2
1
x3
, x = (x1 , x2 , x3 ) R3 , x1 , x2 , x3 > 0
1
b
+ +
1
c
= min{ax + by + cz : 0 < x, y, z,
x+
y+
z = 1}.
x+ y+ z = 1,
p 1
1
1
1 1 1
1 = ( ax + by + cz )2 (ax + by + cz)( + + ).
a b c
a
c
b
Hence,
ax + by + cz,
+ 1b + 1c
x=
a2 ( a1
1
1
1
.
1
1 2, y = 2 1
1
1 2, z = 2 1
+ b + c)
b (a + b + c)
c ( a + 1b + 1c )2
a
b
c
= = .
d
e
f
x=
and
y=
z=
a( a1
1
1
1
1
1 =
1
1 =
1
1
+ b + c)
d( d + e + f )
(a + d)( a+d +
1
b( a1
1
b
+ +
1
)
c
1
c( a1
1
b
+ +
1
)
c
1
b+e
,
1
)
c+f
e( d1
+ +
1
+
1
b+e
,
1
)
c+f
f ( d1
1
1
1
1 =
1
+ e + f)
(c + f )( a+d +
1
b+e
.
1
)
c+f
1
e
1
)
f
a
b
c
= = .
d
e
f
(b +
1
e)( a+d
Exercises
1. Suppose a, b > 0. Prove that
a + g2 (a, b)
a+b
g2 (a,
),
2
2
with equality iff a = b.
2. State and prove an analogue of Theorem 7 and its corollary for h2 , where
h2 (x) =
2
+
1
x1
1
x2
, x = (x1 , x2 ) R2 , xi , x2 > 0.
3
3 abc = min{ax + by + cz : 0 < x, y, z, xyz = 1}.
Deduce Corollary 1.
7. More generally, if xi 0, i = 1, 2, . . . , n, x = (x1 , x2 , . . . , xn ), and
gn (x) =
x1 x2 xn .
Prove that
ngn (x) = min{
n
X
xi yi : y1 y2 yn = 1, yi > 0, i = 1, 2 . . . , n}.
i=1
a+b a+b+c
,
),
2
3
a+b a+b+c
,
),
2
3